Google AI Studio Vibe Coding: How to Build a Free App
Garrett Pierson
Google AI Studio’s Build mode lets you describe an app in plain English and get a working, deployable version back — in a browser, for free, with no credit card. Google’s product name for it is Build mode. Google’s own blog calls what you do in it vibe coding.
If you’ve been sitting on an app idea, this is the cheapest place to find out whether it works. The other AI app builders do have free tiers — Lovable gives you five credits a day — but they meter you by the prompt. Google gives you a full development environment and somewhere to put the finished thing.
Before you start
The full list of what you need:
- A Google account. That’s it. No card, no Cloud project.
- About thirty minutes for a first working version.
- One clear idea of what the app does and who uses it.
Build mode is free within a daily quota, and Google deliberately doesn’t publish a number for it — the allowance shown inside your own account is the real one. The code it writes is yours to keep, too: GitHub sync and a ZIP download are both built in, which is more than no-code platforms will give you.
Step 1 — Open Build mode
Go to aistudio.google.com and click the Build tab. Google sets up your GEMINI_API_KEY behind the scenes, so there’s no key to copy or paste anywhere.
If you want to see it work before you commit your own idea to it, hit the I’m Feeling Lucky button. It writes a project prompt for you; run that prompt and you’ll watch the whole loop end to end in about a minute.
Done looks like: an empty prompt box and a preview panel, side by side.
Step 2 — Describe the app in one paragraph
Write who uses it, what screens it has, and what happens when someone clicks. Concrete beats clever:
“A tip tracker for restaurant servers. One screen to enter a shift: date, hours worked, cash tips, card tips. A list of past shifts underneath. A summary at the top showing this week’s total and average hourly. Save everything on the device.”
That’s the whole skill. A vague description produces a vague app, which is the single biggest reason people bounce off these tools. We cover how to write these descriptions properly in our guide to vibe coding.
Done looks like: a paragraph naming one user, three or four screens, and what each button does.
Step 3 — Watch it build, then click through it
The agent writes the files, installs what it needs, and shows you a live preview. Then you use the app like a customer would — enter real numbers, click every button, try to break it.
Write down what’s wrong as you go. Don’t fix things one at a time while you’re still exploring; you’ll waste generations.
Done looks like: a list of five to ten specific complaints.
Step 4 — Fix things with annotation mode
Annotation mode is the feature that makes Build mode faster than typing prompts. Click or highlight the part of the screen that’s wrong and describe the change right there. You never have to explain which button you mean or hunt through files for it.
Work through your list. Each fix takes under a minute.
Done looks like: you click through the app again and can’t find anything to complain about.
Step 5 — Add a backend when you need one
Web apps in Build mode get a React frontend and a Node.js server runtime, so anything that needs to stay secret stays on the server. Angular and Next.js are supported too, with React as the default.
From plain English you can ask it to:
- Install npm packages
- Set up Firebase Authentication and Cloud Firestore for logins and a real database (it asks you to approve first)
- Call Gemini for the AI features inside your app
- Connect Gmail, Sheets, Docs, Drive, Calendar, or Maps — AI Studio handles the Google Workspace OAuth setup and stores the credentials server-side
Building for phones instead? Build mode also generates native Android apps in Kotlin and Jetpack Compose, previews them in a browser emulator, and installs them on a plugged-in device.
Step 6 — Deploy it
Publishing gets your app a live HTTPS URL. Deploy it as a Cloud Run service, or publish it inside AI Studio on a globally unique your-name.ai.studio address.
The Google Cloud Starter Tier lets you deploy up to two services in a single Cloud Run region without creating a billing account. Past that, you link a real Google Cloud project and pay Cloud Run’s normal rates.
There’s also a share link. Apps are private until you share one, and anyone you share with can read the source and fork it. Your API key stays server-side, but their usage runs on your quota and your bill.
Step 7 — Get the code out
Three exits, all built in:
- GitHub sync — link a repo, push with AI-written commit messages, pull outside changes back in, resolve conflicts in a side-by-side diff.
- ZIP download — self-host it anywhere. You set the
GEMINI_API_KEYenvironment variable yourself. - Antigravity — hand the project to Google’s agentic coding platform and keep working locally.
What “free” actually covers
Three separate meters run at once, and confusing them is where people get surprise bills:
| Meter | What it covers | The free allowance |
|---|---|---|
| AI Studio quota | Your prompts and generations inside Build mode | A modest daily quota. Google publishes no fixed number — check the rate limits page in your own account. Resets daily. |
| Gemini API | Calls your finished app makes to Gemini | Free within your project’s assigned rate limits. Over the limit, requests start failing with a rate-limit error (usually HTTP 429) until the window resets — per-minute limits clear in a minute, daily ones at midnight Pacific. |
| Cloud Run | Hosting the deployed app | Two services in one region on the Starter Tier, no billing account required. |
Two more things before you build something serious:
- Free-tier content can be used to improve Google’s products. Paid-tier content isn’t, under Google’s paid-services terms. Don’t feed the free tier anything confidential.
- Paid usage is prepaid up front. Going paid means linking a Cloud Billing account and normally prepaying at least $10. Run out of free quota and the project stops working until the quota resets.
Paid Google AI Pro and Ultra subscriptions raise your quota inside the AI Studio interface. Neither one is an API plan — an app you deploy is still metered and billed on its own, so check the current price on Google’s plans page before you assume a subscription covers your app’s traffic.
What trips people up
- Sharing spends your quota. Every call a person you shared with makes runs on your key and your bill. A share link that catches on can drain a free allowance in an afternoon.
- One agent turn costs many prompts. The coding agent reads files, runs code, and thinks in loops. Google says an unusually complex interaction can consume three to five million tokens and cost up to about $5 on paid usage.
- The model underneath moves. Build mode currently runs on Google’s Antigravity coding agent, which defaults to Gemini 3.7 Flash. Google doesn’t promise that stays fixed, so treat any tutorial’s model-specific advice as a snapshot.
- “Production-ready” is Google’s marketing language. You own what the app does: the security holes, and whatever the law says about the data you’re storing. Read the code before real people’s data goes into it.
- Free hosting stops at two services. That covers a prototype and a demo. A real launch means a billed Cloud project.
The bottom line
Google AI Studio is the lowest-friction on-ramp in vibe coding right now: a Google account, a paragraph of English, and thirty minutes gets you a real app on a real URL. If you want to compare it against Lovable, Bolt, and the rest before committing, we ranked them in the free AI app builders roundup and the full builder comparison.
Building the app was always the easy half. Choosing an idea people will pay for, pricing it, and putting it in front of the right people is the other half — and that’s what Software Secrets 2.0 is for. The whole book is free.